• Home
  
  
• Übersicht
• Forschung
• Lehre
• Mitarbeitende
• Publikationen
• Auszeichnungen
• Sonstiges
  
  
• Aktuelles
• Kontakt

GET-Forschungsseminar Abstracts

Hierarchical Color Segmentation for Region-Based Visual Attention

Yuan Gao, GET Lab

Vortrag: Mi. 06.06.2012, 16:30, Raum P 1.4.17

Zusammenfassung:

In control of mobile robots, computer vision plays an important role. The current saliency detecting approach used in GET Lab performs color-segmentation as an initial step and then determines saliency using region lists. This work extends the existing implementation by enabling support for hierarchical region-based segmentation. The input image is segmented in different levels. At first the image is segmented into a few regions with coarse granularity. In the next step each region produced is segmented in a finer granularity, so that the big regions are split into smaller ones. The process continues until a predefined level is reached, while all the parent-child (region-subregion) relations are stored as a region-tree. Experiments are conducted to test this approach applied to attention-related problems: Time pressure is simulated by limiting processing to certain levels; Fast scene classification based on the "Gist" obtained from low grain regions; and extracting objects by backtracking from salient subregions to parent regions.